This paper examines some common problems in Human-Robot Interaction (HRI) causing failures and troubles in Chat. A given use case's design decisions start with the suitable robot, the suitable chatting model, identifying common problems that cause failures, identifying potential solutions, and planning continuous improvement. In conclusion, it is recommended to use a closed-loop control algorithm that guides the use of trained Artificial Intelligence (AI) pre-trained models and provides vocabulary filtering, re-train batched models on new datasets, learn online from data streams, and/or use reinforcement learning models to self-update the trained models and reduce errors.

翻译：本文探讨了人机交互（HRI）中导致聊天失败与故障的一些常见问题。针对特定用例的设计决策始于选择合适的机器人、合适的聊天模型、识别导致失败的常见问题、确定潜在解决方案以及规划持续改进。总之，建议采用闭环控制算法，指导使用训练好的人工智能（AI）预训练模型，并提供词汇过滤功能，在新数据集上重新训练批量模型，从数据流中在线学习，以及/或使用强化学习模型自我更新训练模型以减少错误。